Uncle Sam's next big super might not use GPUs

Posted by NamelessVegetable@reddit | hardware | View on Reddit | 6 comments

[-]

EmergencyCucumber905@reddit

Sounds neat. So instead of a stream of sequential instructions the program is a dependency graph.

[-]

nanonan@reddit

Neat in principle, a headache in practice.

[-]

EmergencyCucumber905@reddit

Why? Processors with out of order execution already do something similar internally.

[-]

With AMD providing suitable FP64 GPUs such as MI430X and its successors, my expectation is that any future HPC system is going to go to them. The DoE is willing to take some risks, but I just can't see them going with something as unproven as a large scale dataflow processor cluster. Even when GPUs were the new kid on the block, large SIMDs were already well understood.

Also, "our compiler can make this weird architecture work" has a terrible track record in practice.

Rather than trying to port workloads to run on its chips, NextSilicon has built a compiler that it claims allows it to run any existing C, Python, Fortran, or CUDA codebases on its chips. As we understand it, it works by initially running these workloads on the CPU. The compiler then captures the compute graph, maps it to the chips, and then optimizes it to maximize performance.

Though I'd be delighted to be wrong about all of this. Having a new competitor in the field would certainly be a nice change of pace.

[-]

EloquentPinguin@reddit

This is a very nice opportunity for nextsilicon, to demonstrate that their promises are real, and open the door for potentially huge contracts if this goes well. Exciting to see if they are able to bring some fresh ideas to the industry.

[-]

Dear-Regret-9476@reddit

That’s nice, give us the ram prices and Epstein files to go please